The LAILAPS Search Engine - A Feature Model for Relevance Ranking in Life Science Databases

Transcription

1 International Symposium on Integrative Bioinformatics 2010 The LAILAPS Search Engine - A Feature Model for Relevance Ranking in Life Science Databases M Lange, K Spies, C Colmsee, S Flemming, M Klapperstück, U Scholz Leibniz Institute of Plant Genetics and Crop Plant Research (IPK), Germany lange@ipk-gatersleben.de IPK Gatersleben, March 22, 2010

2 Outline Information Retrieval in Live Science The LAILAPS Approach Query Engine & Feature Scoring Relevance Prediction The LAILAPS System 03/22/2010 2

3 Life Science Data Universe 03/22/2010 3

4 Life Science Data Universe Data Domain Entry# Data Domain Entry# Literature (PubMed): 19*10 6 Molecules (PubChem) 62*10 6 DNA-sequences (GenBank): 112*10 6 Data Domain Entry# Protein sequences(uniprot): 11*10 6 Protein functions (OLS, PDB, INTACT, PFAM): 1*10 6 Pathway (KEGG): 0.1*10 6 Genes (KEGG): 5* /22/2010 4

5 Life Science Data Universe chlorophyll synthase words: 1111 lines: 164 words: 1516 lines: entries of chlorophyll synthase on average 1200 words per entry on average 200 lines 571,200 words 95,200 lines 1322 pages A4 03/22/2010 5

6 Life Science Data Universe - Search 03/22/2010 6

7 The Story of Searching Data 1. 37% of life science scientists spending over 80% of working time in front of a computer 2. 47% of all scientist make daily use of search engines Divoli A, Hearst MA, Wooldridge MA., Evidence for showing gene/protein name suggestions in bioscience literature search interfaces., Pac Symp Biocomput. 2008: /22/2010 7

8 The LAILAPS Approach 03/22/2010 8

9 Motivation for LAILAPS platform for information retrieval over isolated databases content sensitive relevance ranking user profiles for relevance estimation estimation of relevance factors by tracking user behavior 03/22/2010 9

10 LAILPAS Search Engine relevance score feature vector hit excerpt & link to data browser phrase and conjunctive multi term queries user login Javascript data feature scores browsing behaviour tracker synonym & database filter chart type relevance rating 03/22/

11 LAILAPS Feature Model 1. attribute in the record 2. database in which the hit was found 3. frequency of hits per attribute and record 4. co-occurrence of query terms 5. keyword surrounding the hit position 6. organism that is reported in the record 7. sequence length 8. text position of the hit in the record structure 9. synonym that produced the hit 03/22/

12 LAILAPS Ranking Workflow query Q = t 1 t n find: using a text index system relevance ( 1 9) : using a feed forward neural network rank: sorting of relevance scores relevance ranking 1 n query result R = (t,d,{(a,p)}) score: using features F 1 F 9 relevance score chlorophyll synthase ('chlorophyll', 'Q9MBA1', {('DE', 83), ('RT',2)} ('synthase', 'Q9MBA1', {('DE', 97)} entries feature scores ( 1 9) score freq = 3 score co-ocurence = 1-(distance/maxdist) * {cor: 1; wrong: 0.3}... N feedforward :(neurons = 20, input =9, output =1, edges = 7 x 4) Q9MBA1 = N(68, 27,71,33,80,0,31,0,0) = /22/

13 Query Engine: Text Index decompose text into textual atoms avoid line breaks in tokens do not decompose number groups keep chars in token _-/*+. ignore chars in token ()[]<>'',:' token delimiter whitspace and =;\t stop words (ignore frequent but useless tokens) standard in natural language but, his, nor, than, was, by, how, not, that, we, can, however, of, the, were more in life sciences product, locus_tag, db_xref, region, source, DB, organism, CDS, xref, interpro, submission, submitted, pfam, binding, table, embl, geneid, genomic, transl, cdd, sequence, pubmed, taxon, IEA, NIH, genbank, data, here synonyms (from databases Brenda, UniProt) Synonyms Relations: Protein names: M. Lange, K. IPK-Gatersleben 09/12/

14 Relevance Prediction 03/22/

15 Relevance Prediction: Confidence Classes confidence class accession curated rank lailaps rank relevant problem Acc1 1 5 no wrong ranking Acc2 2 2 yes Acc3 3 3 yes Acc4-4 - additional hit (e.g. synonm) Acc5 4 5 yes Acc6 5 - no database version Acc7 6 6 yes Acc8 7 7 yes Acc9 8 - no text decomposition Acc yes Acc no wrong ranking 03/22/

16 Relevance Prediction: Training Data 03/22/

17 Relevance Prediction: Training Data 22 protein queries using data retrieval systems 1069 manual ranked database records 3 quality classes (high, medium, low) 03/22/

18 Relevance Prediction: Neural Network Training split of training/validation data: 80/20 training epochs: 500 mean square error: 0.33 type: feed forward architecture: /22/

19 User Ranking Profile: Relevance Rating Explicit rating Implicit rating: tracking browsing behavior using JavaScript clicked result entries clicked entries above, below activity time scroll amount mouse movement lost / got focus text selection 03/22/

20 Relevance Prediction: Workflow Manual Rating Predicted Rating Query Results Training Data Feedback System Relevance Prediction 03/22/

21 The LAILAPS System 03/22/

22 LAILAPS Technical Background 3rd-party Retrieval System (e.g. in house) Color Code - Technology: Tapestry / ispring ORACLE 11g Color Code Software Modules: Frontend HTTP/SQLNet JAVA 1.5, JOONE, SPRING Network Communication HTML Result Browser Text Query Render Engine Search Form RMI Result Ranking SQL Administration Configuration Feedback Tracking DBMS Backend/Logic ORACLE 11g 03/22/

23 LAILAPS Customization database to be indexed synonyms configuration 03/22/

24 LAILAPS Benchmark 03/22/

25 Benchmark of Neuronal Network Ranking Precision in % 03/22/

26 Conclusion 1. LAILAPS as search engine for life science dabases 2. objective subjectiveness learning of human relevance classes 3. cost-reduction of scientific information retrieval # database records estimated effort in person day , > 1 4. user > 250 relevance profiles objective rating hardly not possible 5. deterministic quality of database research 03/22/

27 LAILAPS Project: Contact: Acknowledgement Jinbo Chen (IPK Gatersleben) Mandy Weißbach (IPK Gatersleben) Gregor Haberhauer (BASF) Michael Leps (BASF Plant Science) Jens Stein (BASF Plant Science) Röbbe Wünschiers (FH Mitweidae) 03/22/